[s-cars] Fuzz logic vs adaptive learning

Linus Toy linust at mindspring.com
Mon Jan 26 13:50:23 EST 2004


> -----Original Message-----
> [mailto:s-car-list-bounces at audifans.com] On Behalf Of Kirby Smith
> Sent: Monday, January 26, 2004 6:48 AM
> Subject: Re: [s-cars] Fuzz logic vs adaptive learning
[SNIP] 
> knowing just what functions are embedded in the A'pexi.  A 
> large-valued n-dimensional table is not necessarily more 
> sophisticated than an n-dimensional high-order function.  
[SNIP]

As I understand it, an "n-dimension high-order function" (I learned it as an
"n-th degree polynomial") would be an attempt to model the data (usually
obtained empirically) of an n-dimension table (matrix or vector space).

At this point, I'll stop on the math, as its been way way too long...but I do
have an interesting, related story I heard years ago...

I think most people know the Boeing AWACS plane is used as a command, control &
communications platform for managing airspace--it made lots of news for itself
in recent wars.  Keep in mind the plane was developed in the early-70s, and its
original computer technology (IBM 4Pi) had but 16K-words (16 bit word, I
believe) of core memory.  Performance data for the airplanes (time to speed,
time to altitude, run at different base altitudes, different throttle
positions, etc) was all required in order to perform its
function--specifically, compute intercepts and guide relatively dumb fighter
planes to enemy aircraft.  Test flights of various fighter assets gave large
volumes of performance data...a large, n-dimension table/matrix, if you will.
Much more data than could be stored in 16K.  The solution:  curve-fit a n-th
degree polynomial to model each fighter's performance.  IIRC, many models
required only a 5th-degree polynomial, so instead of thousands of look-up
points, the computer needed to store only the 5 coefficients for the
polynomial, and everything else could then be computed.

FWIW, this story was told to impress on me the practical value of math vs. the
brute-force approach of much of today's programming code.  I wonder what
lessons our car's s/w engineers may have learned or ignored from these earlier
times.

 
  _____  
Linus Toy
Mercer Island, WA 
linust at mindspring.com
  _____  





More information about the S-CAR-List mailing list